WSN for Forest Monitoring to Prevent IllegalLogging

Jozef PapánUniversity of Žilina

Faculty of Management Science and InformaticsUniverzitná 8215/1, 010 26 Žilina, Slovakia

Email: jozef.papan@st.fri.uniza.sk

Matúš Jurecka, Jana PúchyováUniversity of Žilina

Faculty of Management Science and InformaticsUniverzitná 8215/1, 010 26 Žilina, Slovakia

Email: {matus.jurecka, jana.puchyova}@fri.uniza.sk

Abstract—Illegal logging is in these days widespread problem.In this paper we propose the system based on principles ofWSN for monitoring the forest. Acoustic signal processing andevaluation system described in this paper is dealing with thedetection of chainsaw sound with autocorrelation method. This

work is describing first steps in building the integrated system.

I. INTRODUCTION

LOGGING represents in presence environmental problem.Human used to utilize wood, sometimes in nonsense

wasting, so the quantity of forest area is decreasing. Thisproblem is growing with illegal logging without plan. Thishuman behaviour endangers not only economical market,because the costs for producing materials from legal loggingis usually higher, but also it endangers flora and the species ofanimals in danger, that they loosing their natural environment.

In our case, we decided to guard the forest using the WSN.Nodes are spread trough the whole area of guarded forest. Wedecided to use WSN because the reliability of one node isnot sufficient. Other reason is that with WSN we are able toincrease the monitored area and increase the reliability of thesystem with decreasing number of false alarm. For easy andfast recognition of chainsaw we use autocorrelation method.

II. ACOUSTIC SIGNAL PROCESSING

Forest is a very nice and calm place for relax but from theacoustic point of view it is noisy environment with numeroussound sources which are making chainsaw sound detection notso easy. These sources includes birds, wild animals, ambientenvironment sounds (streams, wind) and human producedsounds.

Sound detection can be done using several means of signalprocessing including Fourier analysis or methods of templatematching implemented by for example Dynamic Time Warp-ing algorithm or Hidden Markov Models. The main prioritieswhile designing the integrated sensor network for describedpurpose were:

• low power consumption of sensors. This implies usingsimple and easy to implement algorithms for acousticsignal aquisition, processing and analysis

This work was supported by institutional grants of Faculty of ManagementScience and Informatics and IBM Award 2011 "Wireless Sensor Networksfor traffic monitoring".

• low cost of sensors• small sensor size to prevent rising suspition• very low probability of false alarms• high reliability

Chainsaw sound is produced mainly by internal combustionengine with 6000 - 18000 rpm. This results in acoustic signalwith fundamental frequency f0 = 100 − 300 Hz or periodT0 = 3−10 ms. (T0 = 26−80 samples for sampling frequencyfs = 8 kHz). After considering the properties of acquiredsignal the autocorrelation method of analysis was chosen forevaluation for its simplicity and implementation possibility ininteger. It can be defined as follows [1]:

R(m) =

∞∑

n=−∞

x(n)x(n +m) (1)

where x(n) is analysed signal of length N and R(m) is theautocorrelation sequence of length 2N − 1.

Autocorrelation sequence is odd function with maximumvalue at R(0) witch stands for the energy of analyzed signalx(n). Several local maximums are occured at the indexes m =T0, 2T0, 3T0, . . . where T0 is the period of analyzed signal insamples. Autocorrelation sequence R(m) of chainsaw soundx(n) is shown at “Fig. 1”.

III. SIGNAL PROCESSING IMPLEMENTATION

Acoustic signal acquisited from ambient environment bymicrophone is transferred into digital form using Pulse CodedModulation with sampling frequency fs = 8kHz with 12-bitprecision.

Intensity of acquisited sound signal ensures waking upthe microcontroller using interrupt request. Thereafter timewindow of 1024 sound samples (128ms) is being stored inmemory and analyzed for fundamental frequency presenceusing autocorrelation method. Because of autocorrelation se-quence symetry only positive indexes m are computed.

The higher the index value i the higher the probabilityof engine fundamental frequency presence in analyzed timewindow.

If 3 analyzed time windows in sequence are declared withpresence of engine fundamental frequency, the sensor willsleep until the next global network communication occures

Proceedings of the Federated Conference onComputer Science and Information Systems pp. 809–812

ISBN 978-83-60810-51-4

978-83-60810-51-4/$25.00 c© 2012 IEEE 809

0 200 400 600 800 1000

−1

−0.5

0

0.5

1

n

x(n

)

102 samples

0 400 800 1200 1600 2000−50

0

50

100

150

m

R(m

)

102 samples

Fig. 1. Autocorrelation sequence of analyzed chainsaw acoustic signal

Algorithm 1 Chainsaw voice detection algorithm1: i = 1;2: Coefficient R(0) is used as an referency to set up the

detection algorithm sensitivity;3: ith local maximum for m = 20 . . . 85 of R(m) > R(0)

5 isthen found;

4: if the result of local maximum search is ∅ then

5: break;6: else

7: index of ith local maximum mi is stored in mem-ory(retained);

8: end if

9: Afterwards first mi autocorrelation sequence coefficientsare stripped of, index i is raised by 1 (i + +) and wecontinue with step 2;

and the interrupt requests form the microphone are maskedduring this time.

The information output from single sensor includes thesedata:

• fundamental period (if present in analyzed time windows)• acquisited sound intensity• no. of local maximums found (i value) which stands

for probability of engine fundamental frequency presence(T0 = 3− 10 ms) in analyzed time window.

IV. EXPERIMENTS AND RESULTS

The following experiment was performed with the devel-opment board shown in “Fig. 3” in the real forest environ-ment [2], [3]. This development board was recently used forsimilar application [4]. Sound of chainsaw lasts for 10 seconds5 times from different distance. The Results of chainsaw soundevaluation correctness depending on distance are shown in“Fig. 2” During testing there was no time window evaluatedwith presence of engine fundamental frequency when thisevent has not really occured (no false alarm appeared).

0 20 40 60 80 1000

20

40

60

80

100

distance in [m]

evalu

ation c

orr

ectn

ess in [%

]

Fig. 2. Chainsaw sound evaluation correctness depending on distance

Fig. 3. Node of WSN

V. PROPOSED WSN FOR FOREST MONITORING

We propose the network, where the sensing ranges of nodesare overlapped. The sensing range is lower than communica-tion range, so there is no problem to communicate betweennodes also in situation, when some of nodes can damage oris unable to pass the message.

The sensors in the “Fig. 4” are spaced in that way, that theycover the whole forest and they ensure redundancy, scalabilityand increase reliability of decision. The probability of illegal

810 PROCEEDINGS OF THE FEDCSIS. WROCŁAW, 2012

logging is bigger on the forest peripheries so the density ofnodes is higher there.

Fig. 4. Proposed WSN with sensing range of each node

A. Principle of Data Collection

When we determined the method of communication, weconsidered two methods: broadcast for message transmissionand addressed message transmission. Because the communi-cation is the most energy consuming, we decided to use theaddressed message sending. The means of on-line actualisationsoftware can be the other saving energy method [5]. Addressedmessage sending method of communication will be used inthat way that the number of sending messages will be thelowest and we will avoid the loop. The nodes in networkgenerate the rooted tree of the graph, where the vertices arethe nodes and edges connect the nodes which communicatetogether. The node sends the message to the node which hasthe common edge with this node and the node is in lowerlevel of the tree –it is closer to root (“Fig. 6”). Root is alwaysthe gateway(GW) of the WSN. The tree is announced to eachnode in the period of time.

B. Creating the Tree

If we know which nodes are able to communicate, we knowthat only between these vertices we can put the edge.

So at first we need to know the graph, from which we aregoing to create a tree. This can be done through the broadcastmessages. Before monitoring the forest, when the nodes aremounted on the trees, the gateway sends the broadcast. Thenodes that hear the message, they check, whom they heard.Then they send the broadcast and so on. It is obvious thatwhen one side hears the second, it is possible also vice versa.So each side hears and waits some amount of time. If themessage doesn‘t come, it means that no one else is in thenetwork. When the last nodes do not get the message fromnew node, they make the graph of the network, create the treeand then send the graph through the edges of the tree to theeach node and also to gateway. Gateway can get more types ofgraphs because none of the nodes from the biggest level hasthe whole information about the network. But it is easy to joingraphs together (“Fig. 5”). Thanks to this graph, gateway isalso able to evaluate, which nodes are able to sound the same

noise. Probability that the chainsaw was used in the forest ishigher, if more nodes inform about this situation.

G(WSN) = (V,E) (2)

T (WSN) = (VT , ET );V = VT , ET ⊆ E (3)

Fig. 5. Graph of WSN with transmission range of one node

The tree of the graph is made with Breadth-First Search(“Fig. 6”) [6]. After first step, the gateway is creating thetree according to changes in network. We assume the nodesdo not change their positions, so the communication betweentwo nodes is possible, if it was possible before. But we needto know, whether all nodes are in good condition and ifthey are able to communicate or monitor the area. As it wasmentioned, nodes send the message after some time, whenthey hear something, what can be the saw. But all other nodesdo not send anything. If we want to know, whether the nodeis always up, after some time all nodes sends message, withthe announcement of saw hearing or the empty announcement.Then the gateway knows, which nodes are up and it can createnew tree.

C. Synchronization in WSN

Because the nodes communicate in some time windows, inreality it is necessary to synchronize the nodes. For synchro-nization the Cristian‘s algorithm is used [7]. According to thisalgorithm, each node of WSN is able to synchronize its clock

Fig. 6. WSN tree with levels of the tree

JOZEF PAPAN, MATUS JURECKA, JANA PUCHYOVA: WSN FOR FOREST MONITORING 811

from the central point of synchronization. In our case, thispoint is the gateway of WSN. The main condition is that thegateway responses immediately.

The principle of algorithm is, that the node requests theserver (in this way our gateway) to send its current time. Nodeis counting the time of communication and to the sented time,it adds the half of the time of communication (“Eq. 4”). Weassume that request time is the same as the time of response.

Tsyn = Ts + 1/2 ∗ Tc (4)

where Tsyn.. synchronizing time, Ts.. sented time and Tc..communication time.

We can say, that our WSN is divided by the number of hopsto the gateway – through the levels of the tree. It is obvious,that gateway is synchronized always. In predetermined timeblock, the nodes, which are on one hop from the gatewaytries to synchronize. In other time block, the nodes from otherhop synchronizes, etc. The time for synchronization shouldbe set in that way, that the inaccuracies cannot influencethe synchronization process. That is why we set the time ofsynchronization after 10 communication times and for eachhop level we set the time for synchronization for 2 seconds.

VI. CONCLUSION

An approach for the forest monitoring based on wirelesssensor network to prevent illegal logging was proposed in thispaper. Acoustic signal evaluation and the principals of networknodes communication were proposed with the focus on lowpower consumption and reliability of the system. This is stillfar from being final version of forest monitoring system andthere is lot for us to improve.

REFERENCES

[1] J. G. Proakis and D. G. Manolakis, Digital Signal Processing, New York:MPC, 1992.

[2] J. Papán, Systém pre spracovanie a vyhodnocovanie akustického signálu,

Diploma thesis, University of Žilina, 2012.[3] J. Micek and O. Karpiš, “Wireless sensor networks - design of smart

sensor node,” in International conference on military Technologies ICMT

2011, Brno, pp. 1109–1116.[4] O. Karpiš and J. Micek, “Sniper localization using WSN,” in International

conference on military Technologies ICMT 2011, Brno, pp. 1063–1068.[5] O. Karpiš, “Software actualization in WSN networks,” in Information and

Communication Technologies - International Conference, Žilina, 2012,pp. 51–54.

[6] S. Palúch, Algoritmická teória grafov, University of Žilina, 2008.[7] F. Cristian, “Probabilistic clock synchronization,” Distributed Computing

(Springer), vol. 3, 1989, pp. 146–158.

812 PROCEEDINGS OF THE FEDCSIS. WROCŁAW, 2012